#asahi-dev · 2026-07-07 08:36
sofuschaos_princess: one of the buffers just needed a bit more space, i will tests a few more bit streams before updating#asahi-dev · 2026-07-07 08:08
svenyuka: yeah, that could've been it then. it only reproduces when all the stars align which kinda fits to my memory ordering / stale cache line from speculation theory#asahi-dev · 2026-07-07 07:59
chaos_princessyes, t6000#asahi-dev · 2026-07-07 07:59
chaos_princessi also tried crowd_run_4088X2168_fr60_bd8_odd_size_l51 and got this https://pastebin.com/9yjHnSkA#asahi-dev · 2026-07-07 07:59
sofusnice! on t6000?#asahi-dev · 2026-07-07 07:59
chaos_princesscrowd_run_1080X512_fr30_bd8_8buf_l3 works#asahi-dev · 2026-07-07 07:46
sofusi cant get gstreamer to select the right format if i add more than one, so i disabled that for now#asahi-dev · 2026-07-07 07:45
sofuschaos_princess: i force-pushed a fix, please also enable the DEBUG_INST* in case it does not work, should be unrelated to t8103#asahi-dev · 2026-07-07 07:41
yukaBut I was wrong to think it was the main/only issue#asahi-dev · 2026-07-07 07:41
yukaand I think I ran into the mmu issue a few times while using the switch boot cpu code#asahi-dev · 2026-07-07 07:41
yukaSwitching the boot cpu does involve initializing the mmu on the secondary core#asahi-dev · 2026-07-07 07:38
jannauI think it was unrelated. the issue with switching boot cpu is that it is really only safe just before leaving m1n1 since it doesn't update the status property in the cpu nodes in the ADT#asahi-dev · 2026-07-07 07:35
sven this issue was a weird crash when enabling the MMU on secondary cores #asahi-dev · 2026-07-07 07:34
sveni think that was unrelated #asahi-dev · 2026-07-07 07:33
yukaI will test drive it a bit#asahi-dev · 2026-07-07 07:32
yukaOr was that unreliable for other reasons?#asahi-dev · 2026-07-07 07:31
yukaThen we can maybe revisit switching boot cpu#asahi-dev · 2026-07-07 07:13
svenactually, I’ve spent enough hours debugging this crap. let me just merge it and be done with it :D#asahi-dev · 2026-07-07 07:10
svenack#asahi-dev · 2026-07-07 07:08
yukaoops#asahi-dev · 2026-07-07 06:42
jannausven: option 3 is ok as well#asahi-dev · 2026-07-07 06:42
jannauyuka: apple_pcie_init() is called in per-device context. it's called via struct pcie_ecam_ops.init and is unrelated to module init#asahi-dev · 2026-07-07 06:38
svenI think I’ll just reduce it to dsb sy and leave a comment about the cache thing if this happens again #asahi-dev · 2026-07-07 06:37
jannausven: do what you prefer. both options are fine with me#asahi-dev · 2026-07-07 06:33
sventhe kernel’s head.S also does dc ivac with mmu off #asahi-dev · 2026-07-07 06:32
yukabut if we had code to tear down a port and bring it back up again, I would expect to be able to unbind and bind the driver and have it repeatedly probe successfully#asahi-dev · 2026-07-07 06:32
yukathe port setup being in init() doesn't cause any problems in practise, because we don't implement tearing down and bringing back up the pcie ports#asahi-dev · 2026-07-07 06:30
yukaotherwise my pwrctrl code wouldn't work#asahi-dev · 2026-07-07 06:30
yukareturning -EPROBE_DEFER also works from init#asahi-dev · 2026-07-07 06:29
yukajannau: init is for global structures of the driver/module, probe is for per-device initialization? and the ports are per pcie controller#asahi-dev · 2026-07-07 06:20
jannaudsb sy is probably enough to fix it considering that dmb sy avoided it as well#asahi-dev · 2026-07-07 06:18
jannauI wasn't really expecting different result considering the other cores are sitting in tight loops while mmu_secondary_setup() runs. cache invalidation after enabling mmu and caches felt more natural. has dc ivac even an effect when caches are disabled?#asahi-dev · 2026-07-07 06:05
svenI think it’s really just the dsb sy that‘s enough to fix this #asahi-dev · 2026-07-07 06:04
sventested stack invalidation before and after MMU setup and unsurprisingly both work #asahi-dev · 2026-07-07 06:01
jannauwhy do you think so? Do you want to return -EPROBE_DEFER from it?#asahi-dev · 2026-07-07 05:53
nickchanthe pcie-apple.c is directly derived from corellium codes but setup_port() is in probe in the original corellium pcie driver; it must be changed to init at some point in upstreaming#asahi-dev · 2026-07-07 05:42
yukaright?#asahi-dev · 2026-07-07 05:37
yukaI don't think apple_pcie_setup_port should run in init, it should run in probe?#asahi-dev · 2026-07-07 05:34
yukathere is so much wrong with the code as-is afaict#asahi-dev · 2026-07-07 05:33
yukachaos_princess: or even just parts of it, or the devices on it#asahi-dev · 2026-07-06 23:39
StanfordAppliedCyber[m]maybe i can finally abandon my horrible hacks of my hypervisor#asahi-dev · 2026-07-06 23:39
StanfordAppliedCyber[m]debugging my debugusb (heh) right now, making some progress with yuka's help#asahi-dev · 2026-07-06 23:39
StanfordAppliedCyber[m]anyone using kisd on macos: make sure to launchctl disable system/com.apple.deviceinterfaced and kill any current processes#asahi-dev · 2026-07-06 23:07
chaos_princesswouldn't it require moving init code from m1n1 to kernel#asahi-dev · 2026-07-06 23:07
chaos_princesslike, fully power down the pcie link?#asahi-dev · 2026-07-06 23:07
yukaAlso suspend support would be nice#asahi-dev · 2026-07-06 23:05
yukaBut that's an exisiting issue#asahi-dev · 2026-07-06 23:05
yukaUnloading and loading the driver also crashes hard#asahi-dev · 2026-07-06 23:05
yukaProbably not :D#asahi-dev · 2026-07-06 22:57
chaos_princessis it ok that there is no deinit code?